Expiry is what makes a credential honest

Every credential claims that a person can be trusted with certain work. Without an expiry date, that claim stays on the page long after the work, the system and the person's skills have moved on. Expiry forces the claim to be re-earned or withdrawn. For people who approve AI agents' work, that is what keeps a credential honest.

How does a credential become dishonest without anyone lying?

Nobody forges anything. A reviewer earns a certificate fairly, on a system as it stood at the time. The certificate is accurate on the day it is issued.

Then things change. The agent's own certificate lapses every 90 days and is renewed against fresh cases; its depth on a capability may rise or drop. New verdict types appear in the workspace. The business connects new sources, narrows some limits and relaxes others. The reviewer moves teams, takes a long break, or simply forgets the less common procedures.

A year later, the certificate still says the same thing. It now describes a person who was competent on a system that no longer exists. The credential has not been falsified; it has drifted out of truth. Anyone relying on it is relying on a claim nobody would make today.

That is the core of the argument. A credential is a statement about the present dressed up as a record of the past. Expiry reconciles the two.

Why does this matter more for agent work than for many other skills?

Some skills decay slowly and depend little on any one system. A credential in basic first aid might reasonably last several years. A credential to approve what a specific kind of governed agent prepares depends on things that change quickly: what the agent may do, how it shows its reasoning, what the record looks like, which actions are withheld at the platform floor.

The agent itself is held to this logic. Its authority is earned on evidence, enforced at the moment of action, and it expires. If the agent's authority has a 90-day clock, it is odd for the person approving its work to hold a credential with no clock at all. Credential expiry applied to people follows the same reasoning, set to a longer interval because people's judgement is more stable than a model's behaviour on a narrow task.

Who should set the expiry, and how long should it be?

The credential issuer sets it, and the length should match the rate at which the underlying work changes. Too short, and renewal becomes a constant distraction. Too long, and the credential drifts before anyone notices.

For roles that review agent work day to day, a year is a sensible default. It is long enough that renewal is not a burden and short enough to catch a full cycle of product and process change. For people who lead complex, regulated deployments, where the core judgement changes more slowly but the stakes are higher, a longer validity with a heavier initial assessment can make sense.

The other half of the design is renewal. Renewal should be proportionate to what has changed: an update module on what is new and a short check that it landed. That is very different from repeating the original assessment, and it keeps the cost of honesty low.

What does expiry add to credibility?

Several things make a professional certificate believable: an independent assessor, observed work, a clear standard, and a way to check it at source. The article on what makes a professional certificate in AI agents credible goes through them.

Expiry ties those together over time. An independently assessed credential that never lapses is credible on day one and steadily less so afterwards. Add expiry, and add a place where anyone can see whether it is current, and the credential stays credible for exactly as long as it is valid. That is why a public register makes certification mean something: it shows not just who was certified but who is certified now.

What is the trade-off?

Expiry has costs. People must find time to renew. Managers must track dates. Someone who is excellent at the job but missed a renewal deadline loses standing for a while, which feels unfair to them and inconvenient to the team.

There is also a risk of renewal becoming ritual: a quiz clicked through to restore status, teaching nothing. If that happens, expiry adds administration without honesty.

And some people argue that skill does not expire on a schedule. That is true. Expiry is a proxy, not a measurement. Someone might be fully competent the day after their certificate lapses.

These are fair points. The answer is not to abandon expiry but to design it well. Make renewal teach something real about what has changed, so it is worth doing. Make the dates visible well in advance. And accept that a proxy which occasionally lapses a competent person is far better than one which never lapses anyone. Learning that is refreshed on a cycle tends to last longer, a point taken further in what makes AI agent training stick.

What should you do differently?

If you rely on credentials to grant approval rights:

• Stop accepting credentials without expiry as current evidence. Treat them as background.
• Check status at source, not from a copy, when rights are granted and when you audit.
• Put renewal dates in the operating calendar, next to the agent's own recertification dates.
• Define what lapse means. Usually, approval rights at the certified level pause until renewal; the person keeps working.
• Ask what renewal actually covers. If it teaches nothing about recent change, push the issuer for better.

The guide to AI literacy and certification for people who work with agents shows how expiry fits alongside the other parts of certification.